Mechanical properties of HDPE/oxidized Fischer–Tropsch wax composites: Statistical and computational approximation analysis of ANNs and PRA

Numerous theoretical, empirical, and semi-empirical models have been proposed to predict the mechanical properties of polymer composites; however, their predictions often deviate from experimental observations. To address this limitation, this study employs two optimization approaches: polynomial regression analysis (PRA) and artificial neural networks (ANNs) based on a feedforward backpropagation (FFBP) network to predict the elongation at break, stress at break, and Young’s modulus of high-density polyethylene (HDPE)/oxidized Fischer-Tropsch wax composites. The filler volume fraction and the mechanical properties of the constituent phases were used as input parameters. The ANN models employed the TANSIG-PURELIN transfer function and were trained for 1000 epochs, while PRA utilized modified nonlinear polynomial equations to capture the relationship between filler content and mechanical response. Model performance was evaluated using Standard Error (SE), Multiple R, R 2 , Adjusted R 2 , SSE, SSR, F-statistics, and significance of F. For elongation at break, the optimum PRA model achieved SE = 0.919 and R 2 = 0.999, outperforming the best ANN model (TRAINOSS, SE = 1.049, R 2 = 0.999). For stress at break, PRA yielded SE = 1 × 10 −4 -9 × 10 −4 and R 2 = 0.999, compared with SE = 0.219 and R 2 = 0.982 for the best ANN model. For Young’s modulus, the optimum PRA equation achieved SE = 8.79 × 10 −8 and R 2 = 1.000, whereas the best ANN model produced SE = 0.357 and R 2 = 0.999. ANN predictions generally deviated by <2%, whereas PRA produced nearly error-free predictions, demonstrating its superior accuracy for predicting the mechanical properties of HDPE/oxidized Fischer-Tropsch wax composites.